Open Models Almanac
Methodspost-trainingrollout & RL infrastructure

filed under post-training

SLIME

SLIME is an asynchronous RL infrastructure for post-training LLMs.1 It substantially improves training throughput and efficiency, enabling more fine-grained post-training iterations.2

How this definition was written

Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.

  1. we developed [slime](https://github.com/THUDM/slime), a novel **asynchronous RL infrastructure — GLM-5.2 & GLM-5.1 & GLM-5
  2. substantially improves training throughput and efficiency, enabling more fine-grained post-training iterations — GLM-5.2 & GLM-5.1 & GLM-5

Also called asynchronous RL infrastructure, SLIME (Reinforcement Learning framework).

sources
2
model
1
lab adopt it
1
strongest
used

Documented in

Z.ai (Zhipu AI)

Evidence

3 quotes from 2 sources, one card per model, each opening with its strongest.

How sources treat it
used 3

GLM-5.2

Z.ai (Zhipu AI)adopted2 sources

we developed slime, a novel asynchronous RL infrastructure that substantially improves training throughput and efficiency, enabling more fine-grained post-training iterations.

usedoptimizationin GLM-5Z.ai
2 more from this model

we developed slime, a novel asynchronous RL infrastructure that substantially improves training throughput and efficiency, enabling more fine-grained post-training iterations.

usedoptimizationin GLM-5Z.ai

the very popular RL framework that Z.ai uses (SLIME)

usedpost trainingin GLM-5.2Z.ai
↑